Very-WIP: Better error messages.

src/backend/distributed/shared_library_init.c

@@ -22,6 +22,7 @@
 #include "distributed/citus_nodefuncs.h"
 #include "distributed/connection_management.h"
 #include "distributed/connection_management.h"
+#include "distributed/deadlock.h"
 #include "distributed/maintenanced.h"
 #include "distributed/master_metadata_utility.h"
 #include "distributed/master_protocol.h"
@@ -166,6 +167,8 @@ _PG_init(void)
 	set_rel_pathlist_hook = multi_relation_restriction_hook;
 	set_join_pathlist_hook = multi_join_restriction_hook;
 
+	emit_log_hook = DeadlockLogHook;
+
 	InitializeMaintenanceDaemon();
 
 	/* organize that task tracker is started once server is up */

src/backend/distributed/transaction/deadlock.c

@@ -43,6 +43,7 @@
 
 #include "access/hash.h"
 #include "distributed/connection_management.h"
+#include "distributed/deadlock.h"
 #include "distributed/hash_helpers.h"
 #include "distributed/metadata_cache.h"
 #include "distributed/remote_commands.h"
@@ -56,6 +57,7 @@ typedef struct LockDepNode
 	TmgmtTransactionId transactionId;
 	List *deps;
 	int initial_pid;
+	bool *deadlocked;
 	bool visited;
 } LockDepNode;
 
@@ -188,6 +190,7 @@ this_machine_kills_deadlocks(PG_FUNCTION_ARGS)
 
 		initialNode = LookupDepNode(lockDepNodeHash, &CurBackendData->transactionId);
 		initialNode->initial_pid = curProc->pid;
+		initialNode->deadlocked = &CurBackendData->deadlockKilled;
 	}
 	LWLockRelease(&TmgmtShmemControl->lock);
 
@@ -265,8 +268,7 @@ this_machine_kills_deadlocks(PG_FUNCTION_ARGS)
 				if (visitNode == curNode)
 				{
 					elog(WARNING, "found deadlock, killing: %d", curNode->initial_pid);
-					kill(curNode->initial_pid, SIGINT);
+					*curNode->deadlocked = true;
-					pg_usleep(100000);
 					kill(curNode->initial_pid, SIGINT);
 					PG_RETURN_BOOL(true);
 				}
@@ -924,3 +926,21 @@ LookupDepNode(HTAB *lockDepNodeHash, TmgmtTransactionId *transactionId)
 
 	return node;
 }
+
+
+void
+DeadlockLogHook(ErrorData *edata)
+{
+	if (edata->elevel != ERROR ||
+		edata->sqlerrcode != ERRCODE_QUERY_CANCELED)
+	{
+		return;
+	}
+
+	if (MyTmgmtBackendData->deadlockKilled)
+	{
+		edata->sqlerrcode = ERRCODE_T_R_DEADLOCK_DETECTED;
+		edata->message = "deadlock detected";
+		edata->detail = "Check server log for detail.";
+	}
+}

src/backend/distributed/transaction/transaction_management.c

@@ -79,6 +79,7 @@ assign_distributed_transaction_id(PG_FUNCTION_ARGS)
 	MyTmgmtBackendData->transactionId.nodeId = PG_GETARG_INT64(0);
 	MyTmgmtBackendData->transactionId.transactionId = PG_GETARG_INT64(1);
 	MyTmgmtBackendData->transactionId.timestamp = PG_GETARG_TIMESTAMPTZ(2);
+	MyTmgmtBackendData->deadlockKilled = false;
 
 	PG_RETURN_VOID();
 }
@@ -93,6 +94,7 @@ UnsetDistributedTransactionId(void)
 		MyTmgmtBackendData->transactionId.nodeId = 0;
 		MyTmgmtBackendData->transactionId.transactionId = 0;
 		MyTmgmtBackendData->transactionId.timestamp = 0;
+		MyTmgmtBackendData->deadlockKilled = false;
 	}
 }
 

src/backend/distributed/utils/maintenanced.c

@@ -249,7 +249,7 @@ CitusMaintenanceDaemonMain(Datum main_arg)
 	{
 		int rc;
 		int latchFlags = WL_LATCH_SET | WL_TIMEOUT | WL_POSTMASTER_DEATH;
-		int timeout = 10000; /* wake up at least every so often */
+		int timeout = 1000; /* wake up at least every so often */
 
 		CHECK_FOR_INTERRUPTS();
 

src/include/distributed/deadlock.h

@@ -10,6 +10,11 @@
 #ifndef DEADLOCK_H
 #define DEADLOCK_H
 
+#include "fmgr.h"
+#include "utils/elog.h"
+
 extern Datum this_machine_kills_deadlocks(PG_FUNCTION_ARGS);
 
+extern void DeadlockLogHook(ErrorData *edata);
+
 #endif /* DEADLOCK_H */

src/include/distributed/transaction_management.h

@@ -68,6 +68,7 @@ typedef struct TmgmtBackendData
 {
 	Oid databaseId;
 	TmgmtTransactionId transactionId;
+	bool deadlockKilled;
 } TmgmtBackendData;